* Scheduler patch for HARD_REGNO_CALL_PART_CLOBBERED
@ 2002-11-18 3:16 Richard Sandiford
2002-11-19 15:12 ` Richard Henderson
0 siblings, 1 reply; 2+ messages in thread
From: Richard Sandiford @ 2002-11-18 3:16 UTC (permalink / raw)
To: gcc-patches
If an instruction modifies the clobbered part of a partly
call-clobbered register, the scheduler might still move
that instruction across a call.
Patch tested on sh64-elf. Fixes the attached test case
for -m5-32media. OK to install?
Richard
* sched-deps.c (sched_analyze): Check HARD_REGNO_CALL_PART_CLOBBERED.
Index: sched-deps.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/sched-deps.c,v
retrieving revision 1.48
diff -c -d -p -F^[(a-zA-Z0-9_^#] -r1.48 sched-deps.c
*** sched-deps.c 27 Sep 2002 01:07:52 -0000 1.48
--- sched-deps.c 18 Nov 2002 11:10:25 -0000
*************** sched_analyze (deps, head, tail)
*** 1287,1294 ****
SET_REGNO_REG_SET (reg_pending_sets, i);
SET_REGNO_REG_SET (reg_pending_uses, i);
}
! /* Other call-clobbered hard regs may be clobbered. */
! else if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
SET_REGNO_REG_SET (reg_pending_clobbers, i);
/* We don't know what set of fixed registers might be used
by the function, but it is certain that the stack pointer
--- 1287,1298 ----
SET_REGNO_REG_SET (reg_pending_sets, i);
SET_REGNO_REG_SET (reg_pending_uses, i);
}
! /* Other call-clobbered hard regs may be clobbered.
! Since we only have a choice between 'might be clobbered'
! and 'definitely not clobbered', we must include all
! partly call-clobbered registers here. */
! else if (HARD_REGNO_CALL_PART_CLOBBERED (i, reg_raw_mode[i])
! || TEST_HARD_REG_BIT (regs_invalidated_by_call, i))
SET_REGNO_REG_SET (reg_pending_clobbers, i);
/* We don't know what set of fixed registers might be used
by the function, but it is certain that the stack pointer
*** /dev/null Tue Nov 14 21:44:43 2000
--- testsuite/gcc.c-torture/execute/20021112-5.c Fri Nov 15 16:39:39 2002
***************
*** 0 ****
--- 1,26 ----
+ /* Test whether a partly call-clobbered register will be moved over a call.
+ Although the original test case didn't use any GNUisms, it proved
+ difficult to reduce without the named register extension. */
+ #if __SH64__ == 32
+ #define LOC asm ("r10")
+ #else
+ #define LOC
+ #endif
+
+ unsigned int foo (char *c, unsigned int x, unsigned int y)
+ {
+ register unsigned int z LOC;
+
+ sprintf (c, "%d", x / y);
+ z = x + 1;
+ return z / (y + 1);
+ }
+
+ int main ()
+ {
+ char c[16];
+
+ if (foo (c, ~1U, 4) != (~0U / 5))
+ abort ();
+ exit (0);
+ }
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: Scheduler patch for HARD_REGNO_CALL_PART_CLOBBERED
2002-11-18 3:16 Scheduler patch for HARD_REGNO_CALL_PART_CLOBBERED Richard Sandiford
@ 2002-11-19 15:12 ` Richard Henderson
0 siblings, 0 replies; 2+ messages in thread
From: Richard Henderson @ 2002-11-19 15:12 UTC (permalink / raw)
To: Richard Sandiford; +Cc: gcc-patches
On Mon, Nov 18, 2002 at 11:16:19AM +0000, Richard Sandiford wrote:
> * sched-deps.c (sched_analyze): Check HARD_REGNO_CALL_PART_CLOBBERED.
Ok.
r~
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2002-11-19 23:12 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2002-11-18 3:16 Scheduler patch for HARD_REGNO_CALL_PART_CLOBBERED Richard Sandiford
2002-11-19 15:12 ` Richard Henderson
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).